Verification Based on Hyponymy Hierarchical Characteristics for Web-Based Hyponymy Discovery

摘要

Hyponymy relations are the skeleton of an ontology, which is widely used in information retrieval, natural language processing, etc. Traditional hyponymy construction by domain experts is labor-consuming, and may also suffer from sparseness. With the rapid development of the Internet, automatic hyponymy acquisition from the web has become a hot research topic. However, due to the polysemous terms and casual expressions on the web, a large number of irrelevant or incorrect terms will be inevitably extracted and introduced to the results during the automatic discovering process. Thus the automatic web-based methods will probably fail because of the large number of irrelevant terms. This paper presents a novel approach of web-based hyponymy discovery, where we propose a term verification method based on hyponymy hierarchical characteristics. In this way, irrelevant and incorrect terms can be rejected effectively. The experimental results show that our approach can discover large number of cohesive relations automatically with high precision.